PathwayCCA

PathwayCCA is a framework for pathway-level association analysis between multi-omic driver alterations and gene-set expression programs. We propose this framework that extends regularized Canonical Correlation Analysis (RCCA) and tests whether multiple somatic alterations jointly affect the expression of a given pathway. To address confounding from categorical covariates, we introduce an extension of CCA that supports arbitrary linear equality constraints, thereby enabling control over covariate contributions and preventing variables such as cancer subtypes from dominating. We further develop a novel sampling-based hypothesis test that mitigates the effect of spurious correlations common in high-dimensional data.

The repository provides a complete implementation of Regularized CCA (RCCA), Constrained CCA (CCCA), and the proposed PathwayCCA framework with resampling-based hypothesis test.

PathwayCCA Source Code

pathwaycca/constrained_cca.py:

• Implements ccca, which solves CCA under user-specified linear constraints on one coefficient set as proposed in the paper.
• Provides solve_ccca, a wrapper that runs the solver and returns a structured result object.

pathwaycca/plot_cca.py:

• Contains visualization functions for examining CCA results:
  • plt_var: plots correlation-circle plots that shows canonical loadings of variables (with respect to the symmetric latent canonical variate $$Z=U+V$$).
  • plt_indiv: plots sample projections on canonical variates.
  • plc_cca: a wrapper that can contain both visulization plots.

pathwaycca/significance_test.py

• Implements calculate_p_value_ccca, which performs resampling-based significance testing to assess the statistical strength of canonical correlations.

Example Data and Code

Example Inputs:

input_data/pathway_db/msigdb_H_P53.csv constains an example pathway gene set metadata.

input_data/BRCA_subset/ contains a set of example input data that are small subsets of the preprocessed data from the TCGA-BRCA dataset used in the paper

Example code:

We offer a simple example code for the user to experiment with PathwayCCA in pathwaycca_example_code.ipynb using the provided example input data. Note that the outputs may differ slightly from those reported in the paper because the example input data are only subsets of the original dataset (due to GitHub storage limits).